Re: Feedback: #rdfms-identity-anon-resources

At 03:42 PM 7/12/01 -0400, Frank Manola wrote:
>Graham Klyne wrote:
> >
> > My proposal has gratifyingly generated some responses.
>
>I notice you didn't say "has generated some gratifying responses" :-)

:-)

> > In summary, the arguments _for_ anonymous resources in the model/abstract
> > syntax are:
> > (a) difficulty of creating globally unique genids
> > (b) capturing semantics needed to express queries
> >
> > Are there more?
>
>Actually, I think the main (and fundamental) argument for anonymous
>resources in the model/abstract syntax is to support, in the
>model/abstract syntax, the capability described in the current M&S for
>talking about resources that don't have URIs (that you know of),
>e.g.(P41):
>
>"The individual whose name is Ora Lassila, email <lassila@w3.org>, is
>the creator of http://www.w3.org/Home/Lassila."
>
>the graph for which shows a resource *without a URI*.

Thank you for raising this, which should have been mentioned explicitly in 
my summary.

I overlooked this because I had it in mind that this was really just a 
syntactic issue, adequately addressed by allocation of a unique 
identifier.  But, as you say, it is something that is explicitly discussed 
in M&S.

However, it is not clear to me that this is a requirement on the abstract 
syntax:  the graph (graphical representation) that you mentioned is, in my 
view, another possible surface syntax for RDF, and not one which is really 
amenable to being used as a basis for attaching semantics.  Thus, I think 
it is reasonable to view this case as a counterpart to the XML syntax for 
an anynomous resource, where the user may, as a convenience, omit any form 
of identification of the resource.

If a parser can then allocate a unique identifier for this resource, there 
is no need to represent this specially in the abstract syntax.  This seems 
to me to be the same as happens in Skolemization (absent universal 
quantification), which is consistent with interpretation as existentially 
quantified.

I think this is all consistent with what you say...

>That is, the current M&S says we can express this in RDF.  I interpret
>this as meaning that the model/abstract syntax must be capable of
>expressing this, not just the XML serialization.  It seems to me, then,
>that the choice is whether to (a) explicitly support anonymous resources
>as something other than resources with generated URIs;  (b) clarify this
>part of the spec by saying we really have to generate a URI in this case
>(and have the graph show a *generated* URI, not *no* URI);  (c) delete
>this capability from the spec.  This isn't strictly about either (a) the
>difficulty of creating globally unique genids (although this may be an
>obstacle to doing my case (b) above) or (b) capturing semantics needed
>to express queries.  [Note:  my case (a) doesn't preclude generating
>internal identifiers;  it just precludes them being considered the same
>as URIs (e.g., they might only have local scope);  if we're going to
>insist that they be generated *URIs*, that's my case (b).]
>
>I agree (and have said so before) with a translation into logic (which
>Dan has referenced) that views these as existentially quantified
>variables, i.e., the above example becomes:
>
>"There exists a [resource] x such that name(x, Ora Lassila) and
>email(x,lassila@w3.org) and creator(http://www.w3.org/Home/Lassila, x)"
>
>[I *don't* agree that this has anything to do with queries;  the
>interpretation of an existentially quantified variable in a logical
>assertion, as above, isn't the same as the interpretation of a variable
>to be bound in a query expression, in my opinion].

If we can all agree that queries don't figure (directly) in the problem to 
be addressed, then that matter can be deferred, then we can focus on the 
abstract syntax representation of the existential quantification:
   (a) by some distinct representation of the existentially quantified 
resource;  a name with localized scope seems to be a popular option.
   (b) by generation of a unique URI -- effectively Skolemization
which are the same as your cases (a) and (b) above.

For completeness:  nobody, to my knowledge, has argued for changing the 
spec per your (c).

With (b) there is the problem of generating a unique identifier, mentioned 
previously.

I can appreciate the appeal (a), but I think it does bring some significant 
complications of its own:  how to avoid "interference" between such 
resources when RDF graphs are combined.  Two approaches come to mind:

(1) selective update of identifiers so that they remain unique in the 
enlarged graph

(2) explicit representation of the scope of each identifier

Maybe there are others?

...

Finally, I note that DanC has claimed that cases (a) and (b) don't always 
represent the same information.  In an off-list message, Pat Hayes gave an 
assessment of this which I found particularly clear, and would like to 
repeat here (with permission):

At 02:42 PM 7/5/01 -0700, pat hayes wrote:
>Also, a lot turns on who is doing the skolemizing. The issue we started 
>with was whether Dan asserting an existential had the same content as Dan 
>asserting a skolemised form with a gensym. Answer: yes. They aren't 
>strictly, logically, *equivalent* on a technicality, since the gensym is 
>written a different logical language with a new name in it, but they 
>contain the same information and essentially the same conclusions can be 
>drawn from them. (Strictly: any conclusion that doesnt use the gensym name 
>itself can be drawn equally well from either version.) If all you know 
>about 'genid:xyzzy' is some sentence that I send you, then all you really 
>know is (exists (?x)(<my sentence with' ?x' instead of 'genid:xyzzy'>)), 
>and in fact that is all you can infer, apart from such dumb-ass things as 
>(= genid:xyzzy genid:xyzzy). Now of course you are free to generate 
>another genid of your own, and use that as a skolem form; but then you 
>could do that anyway, if I were to give you the existential. There really 
>isnt any inferential advantage to having the existential form over the 
>skolem form.

#g


------------
Graham Klyne
(GK@ACM.ORG)

Received on Thursday, 12 July 2001 17:05:23 UTC